Antahe · AI brand audit

The belief AI holds about Northwind

Reading Northwind's prior — the opinion ChatGPT, Perplexity, Gemini and Google AI already carry about you: the one adjective they're married to, the distance between what you can prove and what they believe, the source-types that built it, and how to move it. A ruler, not a verdict — what AI believes, and how to change its mind.

Prepared for
Northwind Ltd. — Finance & Ops software
Attn: the leadership team
Report date
28 July 2026 · Field window 14–21 July 2026
Classification
Confidential — prepared solely for Northwind Ltd.; not for redistribution
Method in one line
310 audience questions × 4 engines × 3 independent runs = 3,720 captured answers, decomposed into individual claims, each traced to the answers and sources it came from
Prepared by
Elena Marsh — Lead analyst
Priya Raghavan — Data & methods
Tom Okafor — Technical audit
Evidence standard
Every claim traced to the exact answer and the source it came from; exact prompts shown; facsimile screenshots and clickable sources throughout; full statistical basis in Appendix F
In this report
  1. The Prior at a glance — the one adjective AI is married to, the headline Evidence–Prior Gap, the vital signs
  2. The Association Map — Northwind vs each rival, the adjective AI marries to each
  3. The Evidence–Prior Gap — under-rated (evidence beats reputation) and over-rated (a decaying asset)
  4. Source Trace — the source-types feeding the belief
  5. The Playbook — the ranked source-types to win; which fixes auto-ship vs done-for-you
  6. Supporting context & Track over time — how the prior varies by engine and persona, Technical readiness (can AI read your site?) and Offerings readiness (can AI compare & recommend your product?); the belief as a health metric
  7. Appendices A–H — impact sizing, claim & source ledger, the question set, the 51 checks, method, statistics, glossary, sources index

Contents

Section numbers (§) and page references follow the printed edition of this report. Every entry is a live link in this on-screen edition — locked entries open after you unlock below.

The report
Executive orientation & method chips03 §1The Prior at a glance04 1.0How to read this — a ruler, not a verdict04 1.1The one adjective AI is married to04 1.2The Evidence–Prior Gap, at a glance05 1.3The vital signs — visibility is just one06 §2The Association Maplocked §3The Evidence–Prior Gaplocked 3.1Under-rated — evidence beats reputationlocked 3.2Over-rated — a decaying assetlocked §4Source Tracelocked 4.1The sources behind the belief, by typelocked 4.3How the belief forms — six patternslocked §5The Playbooklocked 5.2Ranked source-types to winlocked 5.3The 90-day sequencelocked Supporting context — how the prior varieslocked Technical readiness — can AI read your site?locked Offerings readiness — can AI compare you?locked By persona · claims & sentimentlocked Track over time — the belief as a metriclocked
Appendices
AImpact sizing — cost of the status quo todaylocked BFull claim & source tablelocked CThe question set — 48 representative of 310locked DTechnical audit — all 51 checks itemisedlocked EMethod & collection protocollocked FStatistical basis — CIs, effective n, κlocked GGlossarylocked HSources index — every URL citedlocked
Evidence exhibits (facsimile screenshots)
Ex.AChatGPT — the "6+ months" claimlocked Ex.BGemini — the "no SOC 2" absence-inferencelocked Ex.CGoogle AI — a healthy SMB answerlocked Ex.DPerplexity — forum-as-survey, with its sourceslocked
Executive orientation · Confidential

What AI believes about Northwind — and how to move it

An AI assistant's answer about your brand isn't a lookup or a calculation — it's a prior: a belief assembled from thousands of pages the model read. This report reads Northwind's prior — the single adjective AI is married to you, the distance between what you can prove and what AI believes, the source-types that built the belief, and the ranked plays that move it. We measure the belief and trace where it came from; we do not adjudicate what is true.

In plain terms

We didn’t just count how often AI mentions you. We read 3,720 real AI answers and reconstructed the belief underneath them: the word AI reaches for when it describes Northwind, where that belief runs ahead of or behind what you can prove, and the exact source-types doing the work. Everything here is descriptive — you say X; AI believes Y; here is the gap, and how to close it.

310 audience questions 4 engines × 3 runs 3,720 answers read Run 14–21 Jul 2026 Unit of analysis: the prior (the belief AI holds) Stance: a ruler, not a verdict (we measure distance, not truth) Noise floor ~5 pts · CIs shown as ± (the margin of certainty)
Part 1 of 5

The Prior at a glance — what AI believes about Northwind

The belief in one screen: the single adjective AI is married to Northwind, the headline Evidence–Prior Gap (where the belief runs ahead of or behind what you can prove), and the vital signs beneath it. Visibility is one of those vital signs — a thermometer, not the diagnosis. Everything is descriptive: we measure the distance between your evidence and AI's belief; we do not decide what is true.

1.0 · How to read this — a ruler, not a verdict
In plain terms

Two things make this report trustworthy. First: we read the belief AI holds about you — the adjective it reaches for and the claims it repeats — not a raw mention count. Second: we never rule on what is true. The “evidence” side of every gap is what you told us you can prove; we simply measure the distance to what AI believes. AI also answers differently each time, so every percentage carries a ± margin, and gaps too small to be real are called ties.

The unit of analysis is the prior — the belief, not the mention. A mention count tells you whether AI talks about you; it cannot tell you what AI believes. So every one of the 3,720 captured answers was read for the belief underneath it — the adjective the engines reach for, and the recurring statements ("implementation takes 6+ months", "no SOC 2", "the affordable pick") — and each was traced to the source-type that produced it.

Evidence is yours; the gap is ours to measure. We do not adjudicate truth. Where this report reads a distance between what Northwind can prove and what AI believes (Part 3), the evidence side is brand-declared — captured at onboarding — and observable signals; the belief side is measured from the corpus. We report the distance: you say X; AI believes Y. Nothing here is a verdict on Northwind.

Every reading comes with its receipts. Four findings are heavy enough that we reproduce the captured answers as facsimile screenshots — Exhibits A–D in the Source Trace (Part 4) — each showing the exact prompt (with its question number from Appendix C), the engine, the capture date and run, the sentence that carries the belief highlighted, and, where the engine displayed citations, the actual clickable sources. Where a source is named anywhere in this report, it is linked; the full numbered index is Appendix H.

Every percentage carries uncertainty. Engines are stochastic (they answer differently each time). We ran every question three times per engine and report ± ranges (95% Wilson intervals, Appendix F). Differences smaller than the ~5-point noise floor (too small to be real) are treated as ties, and comparisons are tagged proven (certain beyond the margin) or indicative (directional). No forward projections — we describe the belief as it stands.

The prior, in one paragraphAsk the engines about Northwind and one word keeps coming back: “cumbersome” — enterprise-heavy, slow to roll out. That belief runs behind your own evidence on nearly every attribute that matters: you onboard in ~6 weeks but AI believes 6+ months; you hold SOC 2 but AI believes you don't; you offer 57 integrations but AI believes about a dozen. The one place the belief runs ahead of the evidence — “the affordable, easy pick” with small businesses — is a decaying asset that will correct as models read your live pricing. Visibility (34/100) is one vital sign of all this, not the story. The story is the belief, where it came from, and the source-types that move it.
1.1 · The one adjective AI is married to
In plain terms

Ask any engine to describe Northwind and it hands you a single identity. Across 3,720 answers the association clusters tightly around one idea: a capable product that is a chore to adopt. This is the word that walks into the room before you do — and, today, it’s invisible to you.

AI is married to Northwind as
“cumbersome”

Enterprise-heavy · slow to implement · hard to roll out. Elicited by asking the engines directly and aggregating across every answer, the association the models return for Northwind converges on a single frame: a good tool that costs you months to adopt. It precedes you into every shortlist and quietly sets the terms of the comparison.

The association cluster, by weight indicative
Cumbersome / slow
dominant
Affordable / SMB pick
secondary
Not enterprise-ready
recurring
Narrow integrations
recurring
Why the adjective is the assetThe single word AI marries to you is the thing you actually own in the AI layer — more legible and more moveable than any share-of-voice number. Part 2 maps it against every rival's word; Part 3 measures how far it sits from your own evidence.
1.2 · The Evidence–Prior Gap, at a glance
In plain terms

The belief runs mostly behind what you can prove. On the attributes that decide deals, what you told us you can prove and what AI believes point in different directions. This is a ruler, not a verdict — you say X; AI believes Y — and the full two-sided read is Part 3.

On this attributeYou say (brand-declared)AI believesRead
Implementation time~6 weeks median (cohort data)"6+ months"Under-rated
Security postureSOC 2 Type II + ISO 27001, held today"does not appear to hold SOC 2"Under-rated
Integrations57 connectors, updated weekly"about a dozen"Under-rated
SMB price / valuepriced at the premium end"the cheap, easy pick"Over-rated
The direction that mattersFour of the five attributes we can measure against your declared evidence are under-rated — AI believes worse of you than you can prove, so you're left out of recommendations you objectively deserve. The one over-rated attribute is pleasant but decaying. Part 3 scores both sides in full; Part 5 turns them into the source-types that close the gap.
1.3 · The vital signs — visibility is just one
In plain terms

Six dials describe the health of the belief. Visibility (34/100) only says AI mentions you sometimes; it is the thermometer, not the diagnosis. The dials that matter more are about what AI believes and who hears it — the claim health, the tone, and the wild unevenness across personas.

34 / 100 0 100
AI Visibility Score
One brand-health vital sign — a weighted composite of the six dials at right. A useful thermometer, not the whole diagnosis; what AI actually perceives, and why, is the rest of this report. Category median is 51; the leader (Competitor A) models at 74.
Presence
41
Named in 41% of the 3,720 answers. Basis: strong SMB recall props this up; enterprise questions barely surface you.
Share of voice
28
Your slice of all brand mentions, indexed. Basis: 18% raw SoV vs Competitor A's 47%.
Sentiment
52
Net tone where you ARE mentioned. Basis: 34% positive vs 16% negative — mid-pack, dragged by two recurring claims.
Claim health
38
Prevalence-weighted read on what's said. Basis: two high-prevalence claims (57% and 22%) do most damage.
Technical
82
42 of 51 machine-readability checks pass. Basis: strong site — but the 3 fails are precisely the expensive ones.
Persona balance
22
How evenly visibility spreads across personas. Basis: 71% with SMBs vs 8% with enterprise — a 63-pt spread.
In plain terms

Think of these six dials as a health chart, not a single grade. Visibility (34/100) just says AI mentions Northwind sometimes, rarely as a confident first pick. The dials that matter more are about perception: what AI actually says about you is driven by a handful of outside sources (Claim health 38 — the prevalence-weighted read on what AI asserts), the tone is mid-pack (Sentiment 52), and it's wildly uneven across your personas (Persona balance 22 — small businesses hear about you constantly, enterprises almost never). Your website itself is technically healthy (Technical 82); the problem is what's being said about you, and by whom.

Northwind — AntaheSample report§1 · p. 06
Part 1 was free · Parts 2–5 + appendices below

You’ve seen the belief. The rest shows how to move it.

Unlock the full 30-page read — the Association Map (your word vs every rival's), the two-sided Evidence–Prior Gap, the Source Trace with captured screenshots, the ranked Playbook of source-types to win, plus the supporting depth and every appendix. Enter your work email to unlock.

One work email. No spam.
Something went wrong — please check your email and try again.
Sample report. Antahe · Antahe